kylebrodeur/ microfactory-node-v2

7 Downloads Updated 1 month ago

LoRA adapter on Gemma 4 E4B for 3D print job planning. The first successful fine-tune after v1 (which parroted a template). This is the non-QAT sibling of the recommended v3 QAT model.

ollama run kylebrodeur/microfactory-node-v2

curl http://localhost:11434/api/chat \
  -d '{
    "model": "kylebrodeur/microfactory-node-v2",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='kylebrodeur/microfactory-node-v2',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'kylebrodeur/microfactory-node-v2',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Models

Name

1 model

Size / Usage

Context

Input

microfactory-node-v2:latest

5.3GB · 128K context window · Text · 1 month ago

microfactory-node-v2:latest

5.3GB

128K

Text

Readme

Microfactory Node: 3D Printer (LoRA v2)

LoRA adapter on Gemma 4 E4B for 3D print job planning. The first successful fine-tune after v1 (which parroted a template). This is the non-QAT sibling of the recommended v3 QAT model.

What it does

Given a part, material, and environment, returns structured Advice JSON: nozzle/bed temp, retraction, fan, risk regions, and reasoning from precedent. A deterministic Spine validates every value. Two named agents keep it honest: O’Brien proposes, La Forge inspects.

Training

Base: google/gemma-4-E4B-it
Method: LoRA r=4, α=8, 1 epoch
Dataset: 180 train / 80 eval, live-generated on Modal A10G
Noisy targets (temperature=0.7) to prevent template memorization

Iteration history

Version	Base	Result
v1	gemma-3-1b-it	❌ Parroted template
v2	gemma-4-E4B-it	✅ Well-Tuned
v3 QAT	gemma-4-E4B-it-qat-q4_0	✅ Well-Tuned (QAT, recommended)

Safety

This adapter proposes settings. It does not validate them. A deterministic Spine clamps every proposed value against hard material bounds. The LoRA gives the opinion; the Spine has the veto. The human always decides.

Links

LoRA source: kylebrodeur/microfactory-node-lora-v2
GGUFs: kylebrodeur/microfactory-node-gguf
Code: github.com/kylebrodeur/microfactory-node
Live node: node.microfactory.space
Recommended model: ollama run kylebrodeur/microfactory-node-v3-qat

License

Gemma license (inherited from base model).

# Microfactory Node: 3D Printer (LoRA v2)

LoRA adapter on Gemma 4 E4B for 3D print job planning. The first successful fine-tune after v1 (which parroted a template). This is the non-QAT sibling of the recommended v3 QAT model.

## What it does

Given a part, material, and environment, returns structured Advice JSON: nozzle/bed temp, retraction, fan, risk regions, and reasoning from precedent. A deterministic Spine validates every value. Two named agents keep it honest: O'Brien proposes, La Forge inspects.

## Training

- **Base:** `google/gemma-4-E4B-it`
- **Method:** LoRA r=4, α=8, 1 epoch
- **Dataset:** 180 train / 80 eval, live-generated on Modal A10G
- **Noisy targets** (temperature=0.7) to prevent template memorization

## Iteration history

| Version | Base | Result |
|---------|------|--------|
| v1 | gemma-3-1b-it | ❌ Parroted template |
| **v2** | **gemma-4-E4B-it** | **✅ Well-Tuned** |
| v3 QAT | gemma-4-E4B-it-qat-q4_0 | ✅ Well-Tuned (QAT, recommended) |

## Safety

This adapter proposes settings. It does not validate them. A deterministic Spine clamps every proposed value against hard material bounds. The LoRA gives the opinion; the Spine has the veto. The human always decides.

## Links

- **LoRA source:** [kylebrodeur/microfactory-node-lora-v2](https://huggingface.co/kylebrodeur/microfactory-node-lora-v2)
- **GGUFs:** [kylebrodeur/microfactory-node-gguf](https://huggingface.co/kylebrodeur/microfactory-node-gguf)
- **Code:** [github.com/kylebrodeur/microfactory-node](https://github.com/kylebrodeur/microfactory-node)
- **Live node:** [node.microfactory.space](https://node.microfactory.space)
- **Recommended model:** `ollama run kylebrodeur/microfactory-node-v3-qat`

## License

[Gemma license](https://ai.google.dev/gemma/terms) (inherited from base model).

Paste, drop or click to upload images (.png, .jpeg, .jpg, .svg, .gif)